Claude Fable 5.2
Claude Fable 5.2 is the most capable Anthropic model open to every customer, released on Nov 9, 2026 to replace Fable 5.1. It costs US$ 8 per million input tokens and US$ 40 per million output, 20% less than its predecessor and twice Opus 5.5. In Anthropic's table it leads 10 of 12 general tests and loses two: AutomationBench to Gemini 4 Carbon and the AA-Briefcase Elo to Opus 5.5. Mythos 5.2 is the same model without the security and biology safeguards, for verified organizations only.
Claude Fable 5.2 in official numbers
Data published by the maker in its documentation and announcement. The source of each item is at the end of the page.
- Maker
- Anthropic
- API identifier
claude-fable-5-2(Mythos:claude-mythos-5-2)- Release
- Nov 9, 2026
- Input
- US$ 8 per million tokens
- Output
- US$ 40 per million tokens
- Cache read
- US$ 0.20 per million (2.5% of input)
- Cache write
- US$ 10 (5 minutes) or US$ 16 (1 hour)
- Batch API
- US$ 4 input and US$ 20 output
- Fast mode
- US$ 16 and US$ 80, up to 2.5 times faster per Anthropic
- Context window
- 1 million tokens, no surcharge
- Max output
- 128 thousand tokens
- Reliable knowledge
- through June 2026
- Reasoning
- adaptive, always on; default effort high on the API and medium in the apps
- Where it runs
- Anthropic API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS
- Retirement
- not before Nov 9, 2027
Where Claude Fable 5.2 does well, where it falls behind and who it is for
Where it does well
- Agentic coding: 49.1% on FrontierCode 2.0 at xhigh effort, against 45.5% for Opus 5.5 and 42.8% for GPT-6 Astra in Anthropic's table.
- Cost per task: at high effort, 45.3% on FrontierCode 2.0 at US$ 6.72 per task, close to the top Opus 5.5 score (45.5% at US$ 12.82), per the system card.
- Jump over its predecessor: simulated robotics from 27.6% to 64.9% and Remote Labor Index from 17.9% to 41.3%.
- Cheaper than Fable 5.1: US$ 8 and US$ 40 against US$ 10 and US$ 50, with output up to 30% faster, according to Anthropic.
- Fewer unnecessary refusals: about 55% fewer blocks than Fable 5.1 in Anthropic's early testing.
Where it falls behind
- Price per token: twice Opus 5.5 on input and output. Only cache reads cost the same, US$ 0.20.
- Process automation: 51.4% on AutomationBench, against 53.9% for Gemini 4 Carbon.
- Office work: 1,785 Elo on AA-Briefcase against 1,800 for Opus 5.5; in the system card it also trails on GDPval-AA v2.1 (1,831 against 1,866).
- Security and biology blocked: penetration testing, exploits and most research biology get an answer from Opus 4.8 or Opus 5. Without the safeguards, only Mythos 5.2, with approval.
- Mandatory retention: requires 30-day data retention and does not run under zero data retention without express authorization from Anthropic.
Who it makes sense for
- Agents that work alone for hours, such as system migrations and code audits, where Opus 5.5 at high effort does not reach the result.
- Analysis that has to get the math right and cross-check many documents into a report, spreadsheet or deck.
- An advisor to a cheaper model: Haiku, Sonnet or Opus consult Fable only on the hard step.
- Long scientific research in the terminal: 74.6% on Terminal-Bench-Science, the top score in the table.
When to pick another model
- Short questions at high volume: from high effort upward it reasons before almost every reply, and Sonnet 5.5 or Haiku 5.5 cost far less.
- Business automation workflows, where Gemini 4 Carbon scored higher (AutomationBench).
- Offensive cybersecurity or research biology tasks: Fable refuses and hands off to an older model.
- Customer data under a zero data retention contract, until the company gets Anthropic's authorization.
- Code that forces a tool with tool_choice or edits earlier messages in the conversation: it needs changes before switching.
What Anthropic published at launch
Fable 5.2 and Mythos 5.2 came out together on Nov 9, 2026. Anthropic says they are the same model: Fable ships with classifiers that block high-risk cybersecurity and most research biology, which is why it is open to everyone; Mythos comes without those safeguards through the company's verification programs.
The price fell 20% from Fable 5.1, and the model gained a fast mode its predecessor did not have. The company itself, however, recommends starting with Opus 5.5 for most work and moving up to Fable when Opus at high effort still falls short.
The table below is the one from the announcement, with each model at its highest available effort. The cybersecurity and protein rows are left out because they show Mythos 5.2, not Fable.
| Fable 5.2 Anthropic | Fable 5.1 Anthropic | Opus 5.5 Anthropic | GPT-6 Astra OpenAI | Gemini 4 Carbon | |
|---|---|---|---|---|---|
| FrontierCode 2.0 agentic coding |
49.1% | 32.4% | 45.5% | 42.8% | 41.1% |
| FrontierSWE v2 agentic coding |
69.9% | 56.3% | 62.3% | 65.5% | 61.2% |
| AutomationBench business workflows |
51.4% | 31.4% | 42.5% | 41.4% | 53.9% |
| Terminal-Bench-Science 0.1 agentic scientific research |
74.6% | 52.6% | 58.7% | 68.1% | 64.8% |
| WANDR agentic search, relaxed score |
82.1% | 68.7% | 72.3% | not measured | not measured |
| OSWorld 2.1 computer use |
89.3% | 83.5% | 87.2% | 79.5% | 81.5% |
| RoboDojo Sim robotics |
64.9% | 27.6% | 39.4% | 29.8% | not measured |
| AA-Briefcase v1.1 knowledge work, rubric |
66.1% | 61.5% | 61.2% | 52.0% | 61.8% |
| AA-Briefcase v1.1 knowledge work, Elo |
1,785 | 1,669 | 1,800 | 1,569 | 1,611 |
| Remote Labor Index knowledge work |
41.3% | 17.9% | 21.3% | 20.8% | not measured |
| Humanity's Last Exam reasoning, with tools |
74.2% | 65.6% | 67.7% | 57.2% | not measured |
| Chartography chart reading |
73.4% | 44.8% | 64.4% | 71.0% | 72.5% |
Figures from the Claude Fable 5.2 announcement, with safeguards on. FrontierCode: Fable 5.2 and 5.1 at xhigh, Gemini 4 Carbon at medium. Remote Labor Index: Fable 5.2 at high. The dot marks the best in each row.
Fable 5.2 or Opus 5.5: price per token against cost per task
By the price list, Fable 5.2 costs twice as much as Opus 5.5. By the system card, the math can flip: on FrontierCode 2.0, high effort scored 45.3% at US$ 6.72 per task, while the best Opus 5.5 score cost US$ 12.82. Xhigh reached 49.1% for US$ 14.19, and max fell to 45.7% at a cost of US$ 44.70.
The difference comes from how many tokens each one spends to finish. Anthropic notes that each figure is a single run with no error margin, and advises starting at high effort, moving to xhigh only where a gain has been measured and keeping max for testing the model's ceiling.
Anyone moving from Fable 5.1 needs to review their code: editing an earlier message invalidates the reasoning with a 400 error on every account, only Fable 5.2 and Mythos 5.2 read Fable 5.2 reasoning, and the older computer use tool is gone. Forcing a tool with tool_choice is still not allowed, as on Fable 5.1.
| Fable 5.2 | Fable 5.1 | Opus 5.5 | Sonnet 5.5 | |
|---|---|---|---|---|
| Input | US$ 8 | US$ 10 | US$ 4 | US$ 2 |
| Output | US$ 40 | US$ 50 | US$ 20 | US$ 10 |
| Cache read | US$ 0.20 | US$ 0.25 | US$ 0.20 | US$ 0.10 |
| Batch, input and output | US$ 4 and 20 | US$ 5 and 25 | US$ 2 and 10 | US$ 1 and 5 |
| Latency, per Anthropic | moderate | slower | moderate | fast |
Anthropic documentation, pricing and models overview, checked on Nov 9, 2026.
In the Claude apps
Max, Ultra and premium Team seats include Fable models in their usage limits. On Pro and standard Team seats, Fable runs on usage credits, billed separately. During launch week, until Nov 16, 2026, Pro includes Fable for up to 50% of the weekly limit.
The help center warns that Fable models use up limits faster than other models. For daily use in the app, Opus 5.5 remains the choice that fits the plan.
Examples of using Claude Fable 5.2
A team of Anthropic's own agents, running an earlier version of Fable 5.2, migrated a production service from Python to Rust, which now starts about 26 times faster. It is the kind of long job for which the company recommends moving up from Opus to Fable.
Oracle reports a query combining country debt, repayment history and regional benchmarks: Fable 5.2 returned the expected value, and two other models' SQL summed incorrectly, producing 5 to 8 times the correct amounts.
With the advisor tool, a Sonnet or Haiku does the bulk of the work and consults Fable only on the hard step. The bill stays close to the cheap model's, and the hard decision goes through the more capable one.
Where Claude Fable 5.2 stands out and where it falls behind, by the numbers
Claude Fable 5.2 has published results on 24 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Feb 12, 2026 and Feb 7, 2027 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
It is not in the bottom half of its generation on any test with 6 or more models compared.
See all 24 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| GPQA Diamond | 93.9% | #17 / 74 (tied with 1) | #17 / 159 | 96.5% Gemini 4 Carbon |
| AIME 2024 and 2025 (OTIS) | 99.7% | #16 / 76 (tied with 1) | #16 / 156 | 100.0% GPT-5.5 |
| Dtbench | 99.2% | #1 / 71 | #1 / 145 | 99.2% Claude Fable 5.2 |
| Lmca | 70.61 | #1 / 71 | #1 / 124 | 70.61 Claude Fable 5.2 |
| CritPt | 35.7% | #2 / 77 | #2 / 112 | 36.0% Gemini 4 Carbon |
| SciCode | 67.7% | #1 / 75 | #1 / 102 | 67.7% Claude Fable 5.2 |
| ARC-AGI 2 | 93.8% | #4 / 48 | #4 / 90 | 95.0% GPT-6 Astra |
| SimpleBench | 94.9% | #1 / 44 | #1 / 90 | 94.9% Claude Fable 5.2 |
| Frontiermath Tiers 1 3 V2 | 95.1% | #1 / 66 | #1 / 89 | 95.1% Claude Fable 5.2 |
| ARC-AGI | 98.5% | #1 / 48 (tied with 7) | #1 / 88 | 98.5% Claude Fable 5 |
| SimpleQA Verified | 94.0% | #1 / 63 | #1 / 87 | 94.0% Claude Fable 5.2 |
| Chess puzzles | 79.0% | #1 / 63 | #1 / 86 | 79.0% Claude Fable 5.2 |
| Game puzzles | 90.0% | #1 / 61 | #1 / 81 | 90.0% Claude Fable 5.2 |
| Frontiermath Tier 4 V2 | 100.0% | #1 / 57 (tied with 1) | #1 / 71 | 100.0% GPT-6.1 Sol |
| ProofBench | 100.0% | #1 / 60 (tied with 3) | #1 / 71 | 100.0% Claude Fable 5.1 |
| Apex Agents | 80.1% | #3 / 57 | #3 / 60 | 84.2% Gemini 4 Carbon |
| Furniture Assembly | 90.0% | #1 / 35 | #1 / 38 | 90.0% Claude Fable 5.2 |
| Ebr Bench | 90.5% | #1 / 25 | #1 / 31 | 90.5% Claude Fable 5.2 |
| FrontierSWE | 69.9% | #1 / 22 | #1 / 22 | 69.9% Claude Fable 5.2 |
| CursorBench | 60.0% | #2 / 20 | #2 / 20 | 60.4% Gemini 4 Carbon |
| Rli | 41.3% | #1 / 10 | #1 / 17 | 41.3% Claude Fable 5.2 |
| Spireclimb | 19.6% | #4 / 9 | #4 / 9 | 34.0% GPT-6 Astra |
| Spireclimb Sts1 A20 Heart | 21.7% | #2 / 9 | #2 / 9 | 33.3% GPT-6 Astra |
| Spireclimb Sts2 A10 | 17.6% | #4 / 9 | #4 / 9 | 34.7% GPT-6 Astra |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare Claude Fable 5.2 with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Claude Fable 5.2 | Anthropic | 173.44 | n/a | n/a | Nov 9, 2026 |
| Gemini 4 Carbon | Google DeepMind | 169.76 | 1,614 | n/a | Oct 28, 2026 |
| GPT-6 Astra | OpenAI | 169.00 | 1,496 | 82.2 | Sep 3, 2026 |
| GPT-6.1 Sol | OpenAI | 168.48 | 1,500 | 81.6 | Sep 29, 2026 |
| Gemini 4 Argon | Google DeepMind | 166.48 | 1,606 | n/a | Sep 30, 2026 |
| Muse Wonder | Meta AI | 161.68 | 1,537 | n/a | Oct 12, 2026 |
| Kimi K3.1 Pro | Moonshot | 161.37 | 1,521 | n/a | Oct 23, 2026 |
Claude Fable 5.2 or Gemini 4 Carbon?
Gemini 4 Carbon is the model from another lab closest to Claude Fable 5.2 in capability. Across the 17 tests both took, this is the score:
| Test | Claude Fable 5.2 | Gemini 4 Carbon |
|---|---|---|
| Game puzzles | 90.0% | 73.0% |
| SimpleQA Verified | 94.0% | 77.7% |
| Furniture Assembly | 90.0% | 77.5% |
| FrontierSWE | 69.9% | 61.2% |
| SciCode | 67.7% | 64.2% |
| Frontiermath Tier 4 V2 | 100.0% | 95.1% |
| Apex Agents | 80.1% | 84.2% |
| Frontiermath Tiers 1 3 V2 | 95.1% | 90.5% |
| Ebr Bench | 90.5% | 87.6% |
| GPQA Diamond | 93.9% | 96.5% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
What Claude Fable 5.2 costs to deliver each task
Price per token is what is on the price sheet; cost per task is what lands on the invoice, because it depends on how many tokens the model spends. Each point is an effort level, published by whoever ran the test and compiled by Epoch AI. The further up and to the left, the more score per dollar.
CursorBench
The score scale does not start at zero.
Other models from Anthropic
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| Claude Fable 5.2 | Nov 9, 2026 | 173.44 | n/a | n/a | API access |
| Claude Haiku 5.5 | Oct 7, 2026 | 160.43 | 1,454 | 72.1 | API access |
| Claude Sonnet 5.5 | Sep 28, 2026 | 166.56 | 1,532 | 77.8 | API access |
| Claude Opus 5.5 | Sep 22, 2026 | 168.54 | 1,575 | 83.2 | API access |
| Claude Fable 5.1 | Sep 1, 2026 | 166.99 | 1,580 | 83.4 | API access |
| Claude Opus 5 | Jul 24, 2026 | 164.68 | 1,575 | 80.1 | API access |
| Claude Sonnet 5 | Jun 30, 2026 | 156.94 | 1,488 | 76.0 | API access |
| Claude Fable 5 | Jun 9, 2026 | 164.13 | 1,545 | 83.0 | API access |
| Claude Opus 4.8 | May 28, 2026 | 159.94 | 1,508 | 76.2 | API access |
Have you used Claude Fable 5.2? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about Claude Fable 5.2
How much does Claude Fable 5.2 cost?
US$ 8 per million input tokens and US$ 40 per million output, 20% less than Fable 5.1. Cache reads cost US$ 0.20, batch is half price (US$ 4 and US$ 20) and fast mode costs US$ 16 and US$ 80.
What is the difference between Fable 5.2 and Mythos 5.2?
They are the same model at the same price. Fable blocks high-risk cybersecurity and most research biology and is open to everyone. Mythos has no such safeguards and is only released to organizations approved by Anthropic's verification programs.
Is Fable 5.2 better than Opus 5.5?
In Anthropic's table it wins almost every test, but trails Opus 5.5 on the AA-Briefcase Elo and, in the system card, on GDPval-AA. It costs twice as much per token. Anthropic itself recommends starting with Opus 5.5 and moving up to Fable when the results fall short. The independent scores are at the top of this page.
Is Fable 5.1 still worth using?
Fable 5.1 became a legacy model and is still available, but it costs more (US$ 10 and US$ 50) and trails Fable 5.2 on every test in the launch table. Switching requires reviewing code that edits the conversation history or uses the older computer use tool.
Where can I use Fable 5.2?
On the Anthropic API (identifier claude-fable-5-2), Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS, as well as in the Claude apps on plans that include Fable models. Anthropic guarantees it will not be retired before Nov 9, 2027.
Where does Claude Fable 5.2 rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, Claude Fable 5.2 ranks #1 out of 275 models, with an ECI of 173.44. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is Claude Fable 5.2 better than Gemini 4 Carbon?
Across the 17 tests both models took and Epoch AI compiled, Claude Fable 5.2 comes out ahead on 10, behind on 6 and tied on 1. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is Claude Fable 5.2 best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of Claude Fable 5.2 are: SciCode (#1 of 75); Dtbench (#1 of 71); Lmca (#1 of 71). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
- Anthropic, Claude Fable 5.2 and Mythos 5.2 announcement
- Anthropic, Claude Fable 5.2 and Mythos 5.2 system card
- Anthropic, What's new in Claude Fable 5.2 (documentation)
- Anthropic, pricing (documentation)
- Anthropic, models overview (documentation)
- Claude Help Center, Fable 5.2 launch offers
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
Read also: Claude Fable 5.2: what it is, pricing, pros and cons
← See the full ranking